Weekly GitHub Report for Tensorflow

Thank you for subscribing to our weekly newsletter! Each week, we deliver a comprehensive summary of your GitHub project's latest activity right to your inbox, including an overview of your project's issues, pull requests, contributors, and commit activity.

Table of Contents

• I. News
  • 1.1. Recent Version Releases
  • 1.2. Other Noteworthy Updates
• II. Issues
  • 2.1. Top 5 Active Issues
  • 2.2. Top 5 Stale Issues
  • 2.3. Open Issues
  • 2.4. Closed Issues
  • 2.5. Issue Discussion Insights

• III. Pull Requests
  • 3.1. Open Pull Requests
  • 3.2. Closed Pull Requests
  • 3.3. Pull Request Discussion Insights

• IV. Contributors
  • 4.1. Contributors

I. News

1.1 Recent Version Releases:

The current version of this repository is v2.19.0

1.2 Version Information:

Released on March 5, 2025, TensorFlow version 2.19.0 introduces breaking changes to the tf.lite API, including the deprecation of tf.lite.Interpreter in Python with a migration path to ai_edge_litert.interpreter, and updates to C++ API constants for better compatibility. Key improvements include runtime support for the bfloat16 data type in the tfl.Cast operation, alongside the discontinuation of standalone libtensorflow package publishing, while still allowing unpacking from PyPI.

II. Issues

2.1 Top 5 Active Issues:

We consider active issues to be issues that that have been commented on most frequently within the last week. Bot comments are omitted.

1. Issue Replicating TF-Lite Conv2D Quantized Inference Output: This issue concerns a feature request to understand and resolve a consistent mismatch in the output of a quantized Conv2D layer implementation compared to TensorFlow Lite’s output, despite other quantized layers matching perfectly. The user has re-implemented quantized convolution operations using TensorFlow 2.17 and provided detailed code and quantization parameters but observes about a 1% discrepancy in Conv2D outputs that they cannot explain, seeking help to identify the subtle difference in TFLite’s exact implementation.

   • The comments indicate an initial investigation suggesting the problem may stem from inconsistencies in TensorFlow’s internal handling of shapes, API behavior, or execution modes, with recommendations to reproduce the issue minimally, trace execution differences, and implement fixes such as shape validation or kernel updates; the issue is acknowledged as investigated but still pending a fix and further discussion.
   • Number of comments this week: 4

Since there were fewer than 5 open issues, all of the open issues have been listed above.

2.2 Top 5 Stale Issues:

We consider stale issues to be issues that has had no activity within the last 30 days. The team should work together to get these issues resolved and closed as soon as possible.

1. TF-TRT Warning: Could not find TensorRT: This issue describes a problem where TensorFlow on Ubuntu 22.04 cannot detect TensorRT despite having a compatible NVIDIA RTX 3050 Ti GPU and CUDA 12.4 installed, with the user having to manually downgrade the NVIDIA driver from version 550 to 535 to maintain system stability. The user reports repeated attempts to uninstall and reinstall TensorFlow without success and is seeking assistance to resolve the persistent "Could not find TensorRT" warning that appears during GPU configuration.
2. SystemError in tf.ensure_shape and tf.compat.v1.ensure_shape when dtype of shape is tf.uint64 and its value is too large.: This issue reports a bug in TensorFlow where using tf.ensure_shape or tf.compat.v1.ensure_shape with a shape tensor of type tf.uint64 containing very large values close to 2^64 causes a SystemError and OverflowError. Specifically, when such large values are passed in eager execution mode, the functions fail with an internal error related to type checking, indicating improper handling of large unsigned 64-bit integers in shape validation.
3. Feature Request: Integrate different Digital Signal Processing into tf.signal: This issue is a feature request proposing the integration of advanced digital signal processing (DSP) functionalities, similar to those found in the julius library, into TensorFlow's tf.signal module. The requester highlights the current lack of sophisticated audio data augmentation tools within TensorFlow compared to PyTorch and suggests that adding these capabilities would enhance audio model training by enabling native, efficient preprocessing and augmentation workflows.
4. [DOCS] Missing complex input for Round op: This issue reports a documentation bug in TensorFlow where the Round operation is described as supporting complex tensor inputs, but in practice, attempting to use a complex tensor with this operation results in an error, requiring users to round the real and imaginary parts separately. The user provides a reproducible example and error logs showing that the operation fails due to a missing device kernel for complex inputs, highlighting a discrepancy between the official documentation and the actual behavior in TensorFlow version 2.15.0.
5. tf.raw_ops.Unbatch aborts with "Check failed: d < dims()": This issue reports a bug in TensorFlow version 2.17 where the tf.raw_ops.Unbatch operation aborts with a fatal check failure error "Check failed: d < dims()" when invoked with certain tensor inputs. The problem has been reproduced using TensorFlow Nightly on a Linux Ubuntu 20.04.3 LTS system with Python 3.11.8, and it causes the program to crash due to an invalid dimension check in the tensor shape handling code.

2.3 Open Issues

This section lists, groups, and then summarizes issues that were created within the last week in the repository.

Issues Opened This Week: 22

Summarized Issues:

• Quantized Conv2D Implementation Differences: A custom re-implementation of quantized Conv2D in TensorFlow 2.17 produces a consistent ~1% output mismatch compared to the official TFLite Conv2D, despite using the same quantization primitives and matching exactly for DepthwiseConv2D. This indicates a subtle difference in TFLite’s Conv2D implementation that remains unresolved.
• issues/104605

• Timing Discrepancies in CPU Operator Profiling: A custom butil::Timer inside a TensorFlow CPU operator's Compute function reports significantly lower wall time than TensorFlow's full trace profiling for the same operator. This raises questions about timing differences during protobuf deserialization and internal execution.
• issues/104748

• Bazel Build Failures and Workarounds: Bazel fails to build the "tensorflow-lite-select-tf-ops.aar" package on Ubuntu 22.04.4 LTS and Windows with Docker due to ambiguous configurable attribute matches and missing pthread functions. A partial workaround involves adding the "--config=android_arm64" flag, but no complete fix is available.
• issues/104763

• Tensor Shape and Dimension Mismatch Crashes: TensorFlow 2.17 and 2.21.0 exhibit runtime crashes and CHECK failures caused by dimension mismatches in tensor scatter operations, invalid dimension indices in DepthwiseConv2dNativeBackpropInput, and improper tensor shape handling in multi-GPU setups. These errors lead to process-terminating aborts and native C++ CHECK failures.
• issues/104793, issues/104881

• Data Type Mismatch and Core Dumps: Bugs in TensorFlow 2.17 cause core dumps due to data type mismatches, such as passing tf.int32 instead of a tensor for Tsplits in UnicodeDecodeWithOffsets and mixing float32 and float64 types in ResourceApplyAdagradV2 calls. These issues lack graceful error handling and result in crashes.
• issues/104795, issues/104796

• Crashes on CPU-only VMware Virtual Machines: TensorFlow 2.20.0 crashes with native C++ runtime errors on CPU-only VMware VMs running Kali Linux due to attempts to initialize CUDA and improper handling of scalar input shapes in collective operations. These failures occur without Python stack traces or exceptions.
• issues/104830, issues/104845

• TensorShape CHECK Failures Causing Native Crashes: Calling tf.raw_ops.Unbatch() with simple 1-D inputs and tf.raw_ops.BlockLSTMGrad with minimal valid-rank tensors in TensorFlow 2.20.0 triggers fatal C++ CHECK failures in tensor_shape.cc. These errors cause process-terminating aborts without Python-level exceptions.
• issues/104846, issues/104847

• cuSolverDN Initialization and GPU Memory Corruption: TensorFlow 2.17 experiences intermittent core dumps due to cuSolverDN library initialization failures during LU decomposition on multi-GPU systems. Additionally, using extreme parameter values in debugging assertions and FFT operations causes GPU memory corruption and CUDA out-of-memory errors leading to crashes.
• issues/104852, issues/104853, issues/104854

• CUDA Device Context Initialization Failures: Multiple bugs cause TensorFlow to crash during CUDA device context initialization when performing reduction operations on large integer tensors, inverse FFTs with large axis values, and rfft2d calls with special characters in the name parameter. These failures result in out-of-memory errors before computation begins.
• issues/104855, issues/104859, issues/104861

• GPU Memory Corruption and Invalid Resource Handles: TensorFlow 2.10 and 2.17 suffer from CUDA memory corruption, invalid handle errors, and crashes during operations like tf.map_fn with parallel_iterations, lu_solve, sparse operations, and conv3d on GPU-enabled systems. These issues cause failures in cuSolverDN creation and GPU-to-CPU memory transfers.
• issues/104856, issues/104858, issues/104862

• Invalid Axis Parameter Causing GPU Context Failures: Calling tf.raw_ops.GatherV2 with an invalid axis parameter (axis=9) for a 2D tensor causes CUDA device context initialization failure and application crashes due to out-of-memory errors during GPU context acquisition.
• issues/104876

• Documentation Inconsistencies in Python Version Support: TensorFlow documentation inconsistently reports Python version support by mentioning Python 3.12 as the latest supported version, despite official support for Python 3.13. This discrepancy may cause confusion for users regarding compatibility.
• issues/104919

2.4 Closed Issues

This section lists, groups, and then summarizes issues that were closed within the last week in the repository. This section also links the associated pull requests if applicable.

Issues Closed This Week: 4

Summarized Issues:

• GPU Support for Windows TensorFlow: The issue discusses the discontinuation of TensorFlow GPU support for Windows due to technical and maintenance challenges, which affects users needing compatibility with newer versions of Torch and CUDA. This has caused concerns among users looking for recent TensorFlow versions with GPU capabilities on Windows platforms.
• issues/104564

• Issue Reporting and Diagnostic Templates: This issue serves as a template for users to provide necessary system information, TensorFlow version, conversion output, reproducible code, and logs to assist in diagnosing problems with the tflite_convert tool. It lacks specific details but is intended to standardize issue reporting for better troubleshooting.
• issues/104653

• TensorFlow Playground Bug: A bug was reported where the AI model implementation did not visually update the side image during performance assessment, raising doubts about the model's correct execution despite observed improvements in test loss and epochs. This issue highlights potential UI or visualization problems in the TensorFlow playground environment.
• issues/104746

• Data Loader Memory Usage: The issue addresses excessive memory consumption by the data loader when processing large datasets, with a mention of a proposed optimization that was not included. This points to performance inefficiencies in handling large-scale data within TensorFlow workflows.
• issues/104805

2.5 Issue Discussion Insights

This section will analyze the tone and sentiment of discussions within this project's open and closed issues that occurred within the past week. It aims to identify potentially heated exchanges and to maintain a constructive project environment.

1. Performance: optimize data loader for large datasets
   • Toxicity Score: 0.65 (Rapid escalation, aggressive language)
   • This GitHub conversation begins with a user pointing out the absence of an attached proposal and the lack of an issue template, expressing frustration and a dismissive tone that implicitly accuses the original poster of spamming. This initial comment sets a tense atmosphere with a critical and somewhat confrontational sentiment, potentially discouraging constructive dialogue.

III. Pull Requests

3.1 Open Pull Requests

This section provides a summary of pull requests that were opened in the repository over the past week. The top three pull requests with the highest number of commits are highlighted as 'key' pull requests. Other pull requests are grouped based on similar characteristics for easier analysis. Up to 25 pull requests are displayed in this section, while any remaining pull requests beyond this limit are omitted for brevity.

Pull Requests Opened This Week: 4

Key Open Pull Requests

1. CVE-2023-37920 @ requirements_mac.txt: This pull request updates the certifi dependency to version 2023.7.22 or newer in the Mac build environment to address CVE-2023-37920 by removing the untrusted e-Tugra root certificate from the root store, thereby mitigating the risk of Man-in-the-Middle attacks through certificate spoofing.

• URL: pull/104892

• Merged: No

• Associated Commits: 813bf, de857

2. llvm21 aarch64: This pull request proposes updates related to llvm21 for the aarch64 architecture, including modifications to the workspace0.bzl file.

• URL: pull/104910

• Merged: No

• Associated Commits: b5ddf

3. Add missing header inclusion: This pull request adds the missing header inclusion of <stdint.h> which is necessary for the definition of int64_t.

• URL: pull/104927

• Merged: No

• Associated Commits: 833fa

Other Open Pull Requests

• CI Infrastructure and BUILD Configuration: This pull request enhances the BUILD configuration for the requirements updater within the CI infrastructure by introducing explicit visibility settings and CI-specific tags. It also extends timeouts to improve reliability on slower runners and includes minor command formatting cleanup to maintain Bazel best practices without impacting user-facing APIs.
• pull/104931

3.2 Closed Pull Requests

This section provides a summary of pull requests that were closed in the repository over the past week. The top three pull requests with the highest number of commits are highlighted as 'key' pull requests. Other pull requests are grouped based on similar characteristics for easier analysis. Up to 25 pull requests are displayed in this section, while any remaining pull requests beyond this limit are omitted for brevity.

Pull Requests Closed This Week: 4

Key Closed Pull Requests

1. [WIP] Modify CompileFunction to serialize outer dimension markers as JSON: This pull request proposes modifying the XlaCompiler::CompileFunction to serialize collected outer-dimension-dynamic markers into a single structured JSON string stored under the "tf_outer_dim_markers" key in the XLA module proto frontend attributes, replacing the previous per-marker map entries with a compact, dependency-free JSON array format while maintaining backward compatibility and ensuring thread safety.

• URL: pull/104610

• Merged: No

• Associated Commits: b259f

2. formatting with ruff, no code changes: This pull request proposes formatting changes using the tool ruff without modifying any Python code, as a follow-up to a previous pull request, and includes the contributor's CLA signature.

• URL: pull/104639

• Merged: No

• Associated Commits: 03049

3. formatting with ruff, no code changes: This pull request proposes formatting Python code using the tool ruff without making any functional code changes, as a follow-up to a previous pull request, and includes the contributor's CLA signature.

• URL: pull/104640

• Merged: No

• Associated Commits: d4ef6

Other Closed Pull Requests

• Toolchain updates: This topic covers pull requests that update the rules_ml_toolchain to newer versions to support additional architectures or features. These updates ensure compatibility with the latest LLVM versions and improve build support for specific hardware like aarch64.
• pull/104909

3.3 Pull Request Discussion Insights

This section will analyze the tone and sentiment of discussions within this project's open and closed pull requests that occurred within the past week. It aims to identify potentially heated exchanges and to maintain a constructive project environment.

Based on our analysis, there are no instances of toxic discussions in the project's open or closed pull requests from the past week.

IV. Contributors

4.1 Contributors

Active Contributors:

We consider an active contributor in this project to be any contributor who has made at least 1 commit, opened at least 1 issue, created at least 1 pull request, or made more than 2 comments in the last month.

If there are more than 10 active contributors, the list is truncated to the top 10 based on contribution metrics for better clarity.